Modern consumer technologies increasingly rely on cloud services and edge computing to process sensitive user data, which must remain private even during computation. Fully homomorphic encryption (FHE) offers a powerful mechanism for enabling such privacy-preserving computation by allowing data to be processed directly in encrypted form. Since Gentry’s breakthrough, all FHE schemes achieving indistinguishability under chosen-plaintext attacks (IND-CPA) have relied on noise-bearing designs. In these constructions, noise plays a central role in ensuring security, but also introduces substantial efficiency overhead. Thus, whether a secure and efficient noise-free FHE scheme is possible has remained an open question. A recent work marketed as “edge-based quantum-secure FHE’’ (EB-QFHE) claims to resolve this challenge by proposing a noise-free construction over a prime field and asserting both IND-CPA security and improved efficiency (IEEE Transactions on Consumer Electronics, 2025). In this paper, we present polynomial-time key-recovery attacks against EB-QFHE schemes. Our attacks derive two low-degree univariate polynomials from public information, whose unique common root corresponds to the secret key, and recover it by applying standard polynomial root-finding algorithms. Our theoretical analysis shows that the attacks succeed with overwhelming probability. Furthermore, our SageMath implementation demonstrates that the proposed attacks recover the secret key with a 100% success rate in under one second under the parameter settings recommended in the original paper. Our findings have direct implications for consumer electronics security: edge devices, gateways for the Internet of Things (IoT), and smart appliances that adopt EB-QFHE become completely vulnerable, since an adversary can recover the secret key solely from publicly available parameters.
Light field salient object detection (LF SOD) has attracted widespread attention due to its vital role in enhancing the visual perception capabilities of consumer electronics imaging system. However, existing LF SOD approaches fail to fully leverage the complementarity of cross-modal features and the progressive continuity of cross-layer information, potentially causing information loss or redundancy. Besides, they have limitations in modeling long-range feature dependencies, which hinders further improvement in detection performance. To address these issues, in this paper, we develop a novel dual-domain enhanced feature aggregation network (DDEFA-Net) designed to further boost LF SOD performance and facilitate its practical deployment in consumer electronics-oriented vision tasks. Specifically, a dual-domain collaborative refinement strategy is first designed, which effectively integrates the consensus of frequency and spatial domain information, enabling effective multi-scale feature receptive field modeling. After that, to minimize intermodality discrepancies, a bidirectional calibration fusion module is constructed to exploit cross-modal complementary information, facilitating the efficient fusion of all-in-focus features and focal slice features. Finally, a multi-scale feature aggregation module is introduced, which adopts a selective aggregation strategy to align and blend multi-scale features, avoiding feature contamination and enhancing the integrality of saliency prediction. Extensive experimental results on three benchmark datasets show that the proposed DDEFA-Net is superior to the state-of-the-art methods in both quantitative and qualitative comparisons. our code and results are available at https://github.com/wkcbe/DDEFA-Net.
This paper investigates the safety control problem of the autonomous vehicle operating in complex environments with parametric uncertainties and physical constraints. The core challenge lies in achieving a balance between agile obstacle avoidance and the simultaneous satisfaction of velocity and actuator constraints, all of which are sensitive to unknown dynamic parameters. To overcome this issue, an integrated data-driven safety control framework is developed. The proposed framework introduces a special control perspective that transforms external disturbances from adverse effects into beneficial sources of system excitation. This mechanism enables an offline data-driven identification process based on least squares estimation, providing high-precision parameter estimation for vehicle dynamics. Building upon the identified model, a smooth and adaptive artificial potential field (APF) is designed to generate flexible obstacle-avoidance trajectories. To further ensure safety compliance, a control barrier function (CBF)-quadratic programming layer is constructed to handle velocity and actuator constraints. Owing to the proactive smoothness of the APF, the proposed framework mitigates the infeasibility and abrupt switching issues commonly observed in traditional hard-constraint CBF-based methods. Lyapunov analysis guarantees the stability and safety of the closed-loop system. Simulation studies verify the effectiveness and safety of the integrated control framework under complex environmental conditions.
With the rapid development of Unmanned Aerial Vehicle (UAV) technology, 3D reconstruction under embodied consumer-grade UAVs has demonstrated significant application value in fields of smart cities and land surveying. Neural Radiance Fields (NeRF) have become the mainstream approach for novel view synthesis. However, real-world UAV imagery is often constrained to sparse views in complex environments, posing significant challenges for cross-scene generalization. Standard NeRF models struggle under these conditions, as they require dense spatial sampling and fail to generalize to unseen scenes. We propose a generalizable neural rendering network named SparseMamba-NeRF, designed to enhance sparse-view 3D reconstruction quality and achieve robust cross-scene generalization. First, we address feature degradation under sparse views by employing the Global-Local Context-Aware Feature Extraction Network to extract robust 2D geometric priors. Second, we suppress epipolar noise and floater artifacts by using the Spatial-Gated Feature Extractor as an adaptive filter. This prevents overfitting and improves cross-scene generalization. Third, we eliminate the need for dense sampling by using the Ray Mamba module. It models 3D ray propagation as a 1D sequence scan for efficient feature aggregation and color prediction. To evaluate our method, we introduce the sparse-view UAV dataset, which contains largescale aerial scenes with complex terrain. Experiments show that our SparseMamba-NeRF achieves state-of-the-art performance in both per-scene and cross-scene evaluations, maintaining high rendering quality under strict sparse-view constraints.
With the increasing integration of renewable-energy interfaces, energy storage units, electric vehicle chargers, and consumer-side electronic devices, modern power systems face growing frequency-regulation challenges. The dynamic performance of load frequency control (LFC) is essential for ensuring the secure and stable operation of power systems, particularly in the presence of load disturbances. This paper addresses the LFC problem for a single-area power system and proposes a control approach that combines finite-time convergence with deferred prescribed performance. First, a practical finite-time control strategy is proposed, ensuring that system errors are driven into a predefined bounded region within a prescribed time and subsequently converge to an equilibrium neighborhood within a finite time horizon, thereby enabling the preassigned regulation of transient convergence speed and steady-state accuracy. Second, the proposed framework relaxes the conventional requirement on the initial error condition, allowing the control scheme to remain effective even when the initial error lies outside the prescribed bounds. In addition, an online gain regulation mechanism is introduced to reduce the sensitivity of the control performance to fixed gain selection, and an adaptive learning-based update law is employed to regulate control performance under varying operating conditions. Lyapunov-based analysis proves the stability of the closed-loop system and the boundedness of all signals. Simulation results demonstrate the effectiveness and feasibility of the proposed method under parameter variations and external disturbances, indicating its potential for frequency regulation in modern power systems with increasing consumer-side electronic devices and distributed energy resources.
In this paper, the issue of observer-based proportional-integral-derivative (PID) control is investigated for connected and automated vehicles (CAVs) under a dynamic event- triggered protocol (DETP). The platoon dynamics is modeled as a linear discrete-time system with spacing and velocity deviations as the state variables. To address incomplete sensing information as well as communication and computing constraints, an observer-PID control framework is developed from an inte- grated sensing, communication, computing, and control (S3C) perspective. Specifically, the observer estimates unavailable states, while DETP adaptively schedules inter-vehicle transmissions to reduce communication and computational burdens. The objective is to design an observer-PID controller such that the closed-loop system is exponentially mean-square stable and satisfies a prescribed H∞ performance index. By employing Lyapunov stability theory, sufficient conditions are derived in terms of linear matrix inequalities. Finally, simulation results on a six-vehicle platoon are provided to demonstrate the effectiveness of the proposed approach.
Federated learning (FL) enables privacy-preserving collaborative model training on resource-constrained intelligent wearable devices, making it a core technical paradigm for personal health monitoring services. However, traditional reputation evaluation approaches rely heavily on model accuracy while neglecting substantial hardware gaps and variations in data values across wearable devices. As a result, low-spec devices get their contributions unfairly underrated, dampening their participation enthusiasm. To tackle this challenge, we propose a fairness-aware participant reputation evaluation (F-PRE) mechanism for FL in intelligent wearable health monitoring. First, we construct a hierarchical evaluation framework that classifies devices based on inherent hardware constraints and set reasonable basic contribution weights. Both instantaneous and long-term fairness metrics are defined to comprehensively measure real-time training performance and sustained participation value of heterogeneous devices. In addition, we establish a lightweight multi-dimensional reputation model that considers data quality, training stability, participation activity, energy consumption, and collaboration compatibility, combined with peer assessment results to achieve comprehensive scoring. On this basis, we further integrate dynamic deviation calibration and fair incentive strategies into the F-PRE framework to eliminate evaluation bias caused by device performance gaps. Simulation results demonstrate that our F-PRE mechanism reduces reputation evaluation bias for resource-limited wearable devices by approximately 35% on average, effectively balancing the contribution valuation of heterogeneous participants and promoting stable collaboration.
Reliable QR code detection is essential because QR codes are widely used in payments, logistics, authentication, and many other digital services. Inaccurate detection may compromise data integrity, interrupt service workflows, increase manual intervention, and reduce overall system reliability. In practice, QR code detection remains challenging under uneven illumination, partial occlusions, reflective surfaces, and dense multi-instance scenes. In addition, practical deployment requires efficient adaptation to newly introduced QR categories without retraining the entire model. To address these challenges, we propose TASM-Net, a dual-backbone framework for deployment-oriented QR code detection. The framework combines efficient local spatial perception with global structural modeling. A Task-Aware Prompted Adaptation (TAPA) module supports continual category expansion, while a Structure-Aware Feature Modulation (SFM) module improves the coordination of local and global features under structural guidance. To provide an objective and interoperable system-level assessment, we further define a figure of merit (FOM) and develop an IEEE 2668-compliant evaluation scheme based on five attributes: mAP@0.5, F1, FPS, Trainable Params, and GFLOPs. Experimental results demonstrate that TASM-Net achieves 96.1% mAP@0.5 at 94 FPS, providing a favorable trade-off between detection accuracy and inference efficiency while achieving the best overall performance under the proposed IEEE 2668-compliant evaluation framework.
The Internet of Medical Things (IoMT) interconnects medical-grade devices, wearable sensors, and healthcare terminals to enable distributed intelligent diagnosis and remote monitoring. However, the wide heterogeneity of IoMT devices, unstable network conditions, and privacy-sensitive clinical data make traditional Federated Learning frameworks prone to inefficiency and instability. This paper proposes the Federated Efficient Gradient-aware Coordination via Multi-stage Compression in FL (FEGCM-FL), designed for efficient and reliable federated learning in IoMT. The framework employs a gradient-driven two-stage compression strategy that explicitly integrates Hierarchical Momentum-based Sensitivity Pruning (HMSP) with Gradient-aware Low-rank Decomposition (GLoR). HMSP leverages a sliding momentum mechanism to capture multi-round gradient information for dynamic hierarchical sparsification. GLoR adaptively allocates rank resources based on gradient sensitivity and introduces a tunable rank scaling factor to realize a pruning-guided rank adaptation process between pruning and low-rank decomposition, thereby effectively balancing communication costs and model performance. Experimental results across multiple datasets and model architectures demonstrate that FEGCM-FL significantly reduces communication overhead while maintaining high model accuracy.
The consumer resources recommendation in the edge-cloud is a challenging task for administration and monitoring. The authorization of consumers via typical gateway authentication allows multiple un-authorized users to access the core of resource allocation and scheduling. In the proposed system, a novel framework is discussed to enhance the resource allocation schema by providing Zero Trust Architecture (ZTA) at the user phase. The ZTA approach minimizes the dependency of external usage of user authentication and further a neurosymbolic machine learning model bound with user-dynamics and demand (UDD) technique is evaluated via a API server gateway for predictive pattern extraction of resource demand across multiple (dynamic) users. The process builds a customized large learning model (LLM) via Artificial Intelligence across the servers to assure the learning/training of resource scheduler and allocation. Framework expands the customization with LLM building with ZTA authentication of user demand for dynamic and enhanced resource allocation. The technique has secured a performance matrix of 87.42% in resource demand pattern evaluation with actual resources allocation.
Consumer-Centric Digital Twins (CCDT) require privacy-preserving and personalized intelligence on resource-constrained consumer devices such as smartphones and wearables. To support this setting, we propose Scalable Federated Personalized Aggregation with Posterior Approximation (SFed-PAPA), a scalable personalized federated learning framework that serves as the twin assimilation layer. SFedPAPA models client uncertainty with a Gaussian posterior approximation based on Taylor expansion, empirical Fisher, and KFAC, performs posterior-guided adaptive cohort selection, and generates client-specific weighted aggregates under asynchronous updates. We evaluate SFedPAPA on four healthcare-sensing benchmarks spanning classification and regression, and further analyze asynchronous updates, dynamic client participation, hyperparameter sensitivity, computational overhead, and differential privacy. SFedPAPA achieves 95.73% accuracy on HAR, 79.37% accuracy on WESAD, 2.18 MAE on PT, and 6.93 MAE on PPG-DaLiA, indicating its potential for scalable personalization in consumer-oriented digital twins for Healthcare 5.0.
With the continuous growth of urban parking demand, the difficulty of finding available parking spaces has become a prominent problem, primarily due to improper parking behavior of human drivers and limited parking resources. The rise of autonomous vehicles (AVs) offers new opportunities to alleviate traffic pressure caused by improper parking behavior, with intelligent navigation and autonomous scheduling expected to enhance the efficiency of transportation systems. Meanwhile, by optimizing private parking resources utilization, shared parking helps address the issue of limited parking resources. However, due to mismatches between requested parking durations and available shared parking time, existing shared parking systems often suffer from low utilization and high parking request rejection rates. To address these challenges, this paper proposes a shared parking space (SPS) allocation system with AVs based on time-sharing matching. This approach enables AV to utilize two different SPSs during parking, thereby maximizing the utilization of shared parking time. By adjusting the price of SPSs, the system guides the rational distribution of AVs, while maintaining the total parking cost of users across two-stage allocation. Simulation results show that TSM-SP increases the average SPS utilization by at least 16%, reduces the rejection rate of parking requests by more than 50%, and lowers the average total parking cost of users by at least 16%, compared with baseline methods.
This paper studies the optimal operation of a multi-unmanned aerial vehicle (UAV)-enabled system for consumer-level inspection. The system is centered on the UAV docking station providing communication, computational, and energy support for multiple UAVs performing low-altitude inspection across various task locations within a community-scale area. The system model is constructed based on a virtual-node-based graph, which transforms the complex bidirectional flight network into a unidirectional acyclic graph, simplifying the decision-making model. A complex optimization problem is formulated, jointly optimizing the number of deployed UAVs, task allocation, flight paths, battery swapping, and data offloading for total operational cost minimization. To solve this problem, a two-stage joint optimization framework is proposed. In the first stage, the number of UAVs and specific inspection sequences across the task nodes are determined. In the second stage, by leveraging convex envelope relaxations, the original problem is converted into a tractable mixed-integer convex problem optimizing UAVs’ battery swapping and computational decisions. To efficiently solve the mixed-integer problem in task-intensive scenarios, a penalty-based hierarchical heuristic is proposed to decompose the problem into parallel subproblems, tightening the relaxation and obtaining high-quality suboptimal solutions. Simulation results demonstrate the flexibility of the proposed model for consumer-level inspection scenarios and show that the proposed heuristic achieves high computational efficiency.
The increasing penetration of distributed energy resources and frequent extreme weather events pose significant challenges to distribution networks, directly affecting consumer-side power supply reliability. Existing methods cannot effectively coordinate flexible restoration resources, resulting in limited restoration performance and high computational complexity. To address this issue, this paper proposes a sequential service restoration framework for multi-terminal soft open point (MT-SOP) empowered flexible interconnected distribution networks. The proposed framework jointly optimizes MT-SOP control mode transition, DG start-up, and network reconfiguration in a unified mixed-integer second-order cone programming model. Furthermore, a correlation variable Ks,t is introduced to establish the mapping between reconfiguration stages and time intervals, effectively reducing the number of binary variables and computational complexity. Simulation results on the modified IEEE 102-node network and a real 221-node distribution network demonstrate that the proposed method significantly improves load restoration level and reduces computational time. The proposed framework enhances the resilience and operational efficiency of flexible interconnected distribution networks, providing an effective solution for resilient consumer-oriented power supply in active distribution networks.
The paper addresses the suboptimal fault-tolerant synchronization control of complex networks with actuator failures and dynamic topology. To handle the unmeasurable error state, a filtering-based observer approach is designed to parameterize the system state via filtered input and output signals. An output feedback adaptive dynamic programming learning equation and a policy iteration algorithm are proposed to solve the optimal linear quadratic regulator solution using only measurable output. On this basis, a composite fault-tolerant controller integrating optimized output feedback and fault estimation is constructed to achieve fault-tolerant synchronization. Under this framework, a sparse projected gradient is developed to derive the suboptimal network topology. Sufficient criteria are established to minimize the performance index and ensure uniformly ultimately bounded synchronization performance. Note that the obtained results are suboptimal as the l1-norm sparsity constraint leads to a local optimal solution rather than the global one. Finally, a vehicular platoon example is provided to verify the effectiveness of the proposed approach.
A smart energy community is an integrated multiple consumers energy system that combines distributed energy resources, storage systems and elastic loads, enabling efficient energy utilization and self-sufficiency within a defined area. In the context of energy transition and sustainable development, energy trading among smart users has emerged as a critical approach to enhancing energy efficiency, balancing supply and demand, and reducing carbon emissions. However, due to the uncertain distributed energy resources, complex information exchange, security and privacy issues, research on energy trading among users remains to be further explored, with significant potential for emission reduction and efficiency improvement. This work proposes a novel safe multi-agent deep reinforcement learning algorithm for energy trading among multiple consumers to promote local tradings, reduce cost of users and ensure the safety of energy trading. Specifically, the multiple consumers form a community, in which the energy trading problem is modeled as a reasonable partially observable Markov decision process through incorporating temporal information. Then, a safe multi-agent deterministic proximal policy gradient algorithm (SMA-D2P2G) is developed to assist with training the network and solving the energy trading problem. Finally, the algorithm is evaluated on a real-world dataset under cooperation and competition scenarios, demonstrating its effectiveness in reducing electricity cost and improving energy efficiency.
The latest video coding standard, H.266/VVC, has demonstrated significant improvements in compression efficiency compared to H.265/HEVC. Despite its advanced coding techniques, H.266/VVC still faces challenges in meeting the increasing demand for higher perceptual quality and enhanced compression performance. To address these limitations, we propose MDFI (Multi-Domain Features Integration), a compressed video quality enhancement approach that features a novel Frame-Prediction Feature Transform (FPFT) module to process prediction information. Moreover, MDFI integrates a multi-domain feature fusion strategy that effectively combines spatiotemporal characteristics, cross-frequency representations, and compressed-domain prediction information to enhance decoded video quality. Additionally, we introduce a comprehensive dataset that encompasses uncompressed video sequences, corresponding reconstructed versions at multiple QP levels, and predicted frames generated from H.266/VVC compressed bitstreams, providing essential resources for developing and benchmarking video enhancement approaches. Extensive experiments demonstrate that our MDFI approach achieves superior performance to state-of-the-art methods in both objective metrics and visual quality, effectively mitigating video compression artifacts. The code is available at: https://github.com/dangdinh17/MDFI.git.
LungSage has concordance index score of 0.77 on NSCLC-Radiomics (up from DeepSurv by 0.02), strong accuracy 76.5% for PGD-based attacks with only 1.2% clean accuracy lost. With ε=8 Differential Privacy tested across six independent public datasets comprising more than 12,000 test cases and obtaining 78 ms of inference from current consumer hardware running ARM TrustZone, accuracy remains within ±1.5% of un-privatized scores. All assessments were conducted using a common recurrent state-space architecture. Adversaries attack diagnostic accuracy of consumer health services that use imaging to diagnose tumors from less than 13% without applying security measure, and federated learning must find an effective mechanism to provide safeguards while delivering the level of quality typically found in clinical-use systems. Establishing modes of world model to assist with creating clinical judgments regarding the integration of clinical syndromes (i.e., TCM) with imaging data will support patient outcome. LungSage will deliver a world model-based framework to enable the integration of multimodal fusion (i.e., PET-CT images, electronic health records, and TCM syndromes), sequential treatment reasoning, and an infrastructure for privacy and security provided by latent-space differential privacy, trusted execution environment, and adversarial training inference. We anticipate that the latent-space world models developed to date will provide the basis for proactive threat mitigation and create resilient security for next-generation consumer electronics environments.
We introduce an information-theoretic framework to quantify the intelligence of household appliances as efficiency of uncertainty reduction. Intelligence is decomposed into Perceptual Intelligence (Jperc), measured by mutual information between environmental states and observations, and Decision Intelligence (D), measured by normalized conditional entropy with a Wasserstein penalty on outcome quality. We implement a reproducible pipeline and evaluate three appliance domains (robotic vacuum, smart oven, smart thermostat) across baseline, advanced, and task-optimized models in a high-fidelity simulation. The proposed metrics separate designs by sensor/algorithm capability and align with conventional efficiency outcomes. Aggregating Jperc (normalized) and D into an AIS yields a strong correlation with domain benchmarks (r = 0.95, p < 0.001), indicating predictive validity. The framework offers a principled, decomposable metric for design/standardization of intelligent consumer electronics.
